【作者】 黄万继；

【导师】 张洁琳；

【作者基本信息】 大连理工大学 ， 计算数学， 2011， 硕士

【摘要】 场景分类是计算机视觉领域中的一个基础处理过程,在模式识别、机器学习、图像内容理解、图像检索等中扮演重要角色。在特征提取中,经典的SIFT特征因具有对图像的平移、旋转、缩放、甚至对仿射变换保持不变性且具有良好的显着性和鲁棒性而得到广泛应用。而作为分类器之一,SVM以小样本统计机器学习理论为基础,无需依赖于设计者的经验及先验知识,避免了神经网络实现中的经验成分,而且SVM算法最终转化为凸优化问题,具有全局最优性等诸多优点。本文尝试将SIFT特征算法和SVM结合用于场景图像分类识别中。本文主要内容包括：首先对场景分类、SIFT特征、SVM的研究现状进行了概述,然后重点介绍了提取SIFT特征点和SVM分类算法。然后,基于SIFT特征和非线性SVM提出了场景图像分类系统的理论框架。具体工作包括：场景图像的预处理,即将彩色图像变成灰度图像,同时将图像缩放到相同大小尺寸；提出基于SIFT特征和SVM的场景分类算法,先用SIFT算法得到每个场景中各个图像的SIFT特征点,以向量形式存贮,再以这些特征向量作为SVM的原始输入数据进行训练分类。最后,通过数值实验,一方面比较用不同特征点数目的分类正确率,同时还与传统分类方法进行了效果比较。实验结果表明,本文的基于SIFT特征和SVM场景分类在准确率上高于几种传统的方法。更多还原

【Abstract】 Scene classification is a foundational process in computer vision, and it plays an important role in pattern recognition, machine learning, image content understanding, image retrieve. Classical SIFT feature is an invariant feature to rotation, translation in the field of feature extraction. SIFT Feature is provided with good remarkable and robust. SVM is one of newly and effective classifiers. SVM is based on little sample Statistical Learning Theory and independent on designer’s experience. So it avoids empirical in Artificial Neural Networks. SVM is changed to convex optimize, so it guarantees global optimal. This paper presents SVM scene classification method based on SIFT feature.The main content of this paper includes:first, introduce the current research situation of scene classification, SIFT feature, SVM. Then, based on SIFT feature and SVM, we proposed the theory framework of scene classification. Specific work includes:Firstly, we preprocess the scene images before classification, such as convert the color images to gray images, and scale the images to the same size; Secondly, we present the scene classification algorithm based on SIFT and SVM, SIFT feature points are extracted, they are vectors, and then train SVM by those vectors, At last, recognition work is completed using SIFT feature points and SVM. Our experiment result shows that our algorithm performs better than the one based on geodesic distance.更多还原